Back in the mid-late 1980s I was at IBM Research where I was sort of the proprietor of a bunch of Unix boxes (about 200+, about split evenly between Sun machines and RT-PCs running BSD 4.3 (the ACIS port). I was in the Mathematics department and we called it Mathnet. Rekhter developed the BGP stuff on one of our machines and Zadeck implemented the SSA stuff on one.
When the RS-6000 and AIX 3 were under preparation the AIX folks asked us to help get AIX 3 ready for the real world. They gave us a bunch of RS-6000s (we looked at them and laughed - "what a whale" one of us said ... we called the lab Marineland) and we tried integrating them into our UNIX single (single-ish) system image.
We maintained our (rough) image consistency by rdisting stuff, including administrative junk, from a central set of machines to all of the others.
Well, the AIX systems depended on this database called the ODM (we called it the odious data mangler).
The problem was that the designers of the ODM envisioned themselves as managing a mainframe. The ODM commingled information about the installation (the entire site), the instance (the box itself), and the community (users and stuff). This meant that the database included device identifiers.
We tried rdist-ing the ODM from our central machine to all the rest of our machines. Oops ... lobotomies all around.
I won't bore you with all of the details, but it was a struggle.
At one point I threatened the head of AIX 3 development in Austin, TX that I would post the source code of some major component to rec.humor.funny. He was not my friend.
Anyway, we did end up making a list of horrible things that they absolutely had to fix. And, to their credit, they got a lot of it fixed so that when the RS-6000 and AIX 3 did ship it was not humiliating.
It was probably my first run-in with the pragmatics of shipping real commercial production software.
Marc